# Posts by Crunch3r

 21) (Message 48173) Posted 23 Apr 2011 by Crunch3r Post: Just found those over at Youtube... RCOS Spring 2011 RCOS Spring 2011 - Matt Arsenault - Milkyway@Home http://www.youtube.com/watch?v=yR_Xkm2IITY and http://www.youtube.com/watch?v=B1TgxCMfvGk :p 22) (Message 48172) Posted 23 Apr 2011 by Crunch3r Post: Without using an opti app Cpu performance and credits are poor. Is there any plans to put out better cpu apps? I know that the stock app is pretty much unusable on a CPU. FWIW, it seems that i can reduce crunching times to less than ~5h on a XEON(5365@3 Ghz, core 2 based)... Anyway, i do know that that's far away from what the previous opt app did... I'd be interested to know some crunch times on different CPUs with the stock app (cpu:time:credits), before using a LUT for the INT EXP part... (wich i know for a fact that the previous optimized CPU app used and i allready suggested to Matt for the GPU BUT it'll greatly speed up the stock CPU app as well) 23) (Message 48171) Posted 23 Apr 2011 by Crunch3r Post: FWIW, i don't recommend using your 1090 on AQUA, since it will interfere with MW and slow down your GPU, which is a fact. Besides that, the RAC posted by some that your 1090 could achieve over at AQUA is far from reality. Instead I'd suggest using it over at primegrid using the PPS Sieve app where it will get the best RAC possible. I extreamly second that, Crunch3r. Aqua Roqs will need at least 2 free cores to feed the gpu( as in my case on both my i7 and AMD) taking a big hit on ANY gpu'ing. PPS sieve, on my AMD P-IIX4 9650 avg. 24/30 hours, reguardless you get 4500. Yes, that's what i was talking about. AQUA's ROQS is a dog and has some serious issues. It doesn't utilize all cores while running. On a V8 xeon, it uses less than 4 cores most of the time in the hot loop... (between 49 to 55 % progress) Collatz is a good go-to for cpu. Im thinking of building a cheap i7 960 just for Aqua/Roqs? I like the Quantum computer aspect. By the way, what can I expect frm 2 5870? Stock clock, two differnt rigs. Im using TThrotle at 130 deg. F, runs avg. 75% day 100% at nite. They still finish around 2 min. Collatz on CPU using a 64 bit OS is almost as close to PPS Sieve credit wise and works even better along with GPU apps like MW, having no effect on MW GPU times at all. 24) (Message 48165) Posted 23 Apr 2011 by Crunch3r Post: FWIW, i don't recommend using your 1090 on AQUA, since it will interfere with MW and slow down your GPU, which is a fact. Besides that, the RAC posted by some that your 1090 could achieve over at AQUA is far from reality. Instead I'd suggest using it over at primegrid using the PPS Sieve app where it will get the best RAC possible. 25) (Message 48093) Posted 22 Apr 2011 by Crunch3r Post: BOINC UNITED - Let your Computer be part in scientific experiments. BOINC UNITED is a unique, international team open to all that are interested. Be part of something special! Combining our knowledge, having inspirational talks with our Members and helping out the scientific community by participating in various BOINC Projects. Please make sure to visit us at our BOINC UNITED Homepage and our BOINC UNITED Forum! Make sure to join us here at the project to be part of something special. We can help cure diseases, study global warming, discover pulsars, map the milkyway galaxy and do many other types of very important scientific research that helps making the world a better place for all of us. Together we can make a difference! Visit us at: 26) (Message 48084) Posted 20 Apr 2011 by Crunch3r Post: The problem is not the 3850, It's your CPU that causes the crash. The app is compiled with SSE2 enabled and every time the app tries to calculate the likelihood on the CPU, it crashes. There's nothing you can do about that ATM. 27) (Message 48021) Posted 19 Apr 2011 by Crunch3r Post: I have a ATI4875 (Gigabyte branded) as my primary display and yes the APP Driver set was loaded. I am wonder if the ATI 3200 built into the 780G chipset is the problem, and if short of disabling it completely ( I use it to drive a secondary monitor ) is there a means to tell milkyway not to use it. That would also explain why 1 GPU process seems to run OK while the other GPU work units quickly error out. That is the 4875 is processing correctly and the 3200 errors out over and over quickly consuming all the wu. The \$64 question is that up until the recent server issues both GPUs ran fine, in fact they ran at similar speeds. You can disable the 3200 in boinc to prevent it from trashing more WUs. Shut down boinc and after that, create a file called cc_config.xml (using notepad) with the following content:  1  copy that file to "C:\Documents and Settings\All Users\Application Data\BOINC" start boinc again. 28) (Message 47997) Posted 18 Apr 2011 by Crunch3r Post: With no official feedback was it just a hollow promise? I'd rather wait till the next app release with integrated likelihood calculation on the gpu. That might shave off 5 sec. or so... ;) 29) (Message 47744) Posted 12 Apr 2011 by Crunch3r Post: Mine seems pretty responsive, maybe a touch slower than the old app. Overall, I am pleased so far. PS I don't have the latest/greatest hardware out.It is slightly slower. Most of the slowdown is since the likelihood at the end isn't currently done on the GPU which I'm planning on fixing. But even if you do the likelihood on the GPU, it won't be nearly enough to comepensate the total performance loss of 25% compared to the old app. Adding it all up here across the machines, i roughly lose ~100k credits/day using the new app. 30) (Message 47657) Posted 11 Apr 2011 by Crunch3r Post: BOINC United is a unique, international team open to all that are interested. Combining our knowledge, having inspirational talks with our Members and helping out the scientific community by participating in various BOINC Projects. Be part of something special! Are you ready to make a difference ? Please join us BOINC United 31) (Message 47652) Posted 11 Apr 2011 by Crunch3r Post: WOW..... Whatever they just did here triggered a major malfuntion on my big gun battlecruiser!! Not only did it trash all the MW when I restarted BOINC, but the whole load went down the drain!! OUCH!! :-D All I did was stop BOINC to go do a race in iRacing. When I restarted BOINC after the race, all hell broke loose! If that one -> http://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=88232 is your "battlecruiser", then it 's fairly obvious. 6.10.60 Incorrect function. (0x1) - exit code 1 (0x1) milkywayathome_client separation 0.57 Windows x86 double CAL++ Found 3 CAL devices Chose device 0 Device target: CAL_TARGET_610 Revision: 19 CAL Version: 1.4.1332 Compute shader: CAL_FALSE Engine clock: 650 Mhz Memory clock: 1000 Mhz GPU RAM: 341 Wavefront size: 64 Double precision: CAL_FALSE Compute shader: CAL_FALSE Number SIMD: 2 Number shader engines: 1 Pitch alignment: 256 Surface alignment: 256 Max size 2D: { 8192, 8192 } Device does not support double precision Device failed capability check Failed to setup CAL 22:22:17 (4536): called boinc_finish There are plenty of other like that tasks as well. 32) (Message 47519) Posted 10 Apr 2011 by Crunch3r Post: I fixed this last night (https://github.com/Milkyway-at-home/milkywayathome_client/commit/b83566d399c5b63f717d864a3a45b8a04ff2fa9b), so CPU usage should be normal when I update it. I'm going to try fixing a few more problems before I do that though. Yes, works like a charm and CPU usage is even lower (1%) than what the old app used to be. 33) (Message 47383) Posted 10 Apr 2011 by Crunch3r Post: I have tried stock, with an app_info referencing the brook.dll file and with an app_info referencing both the brook32 and brook64 dll files and they all crash out. So far it looks like people with 32-bit systems are running file while those of us with 64-bit systems are not.That's strange since I built it for 32 bit on a 64 bit system. I also didn't use Brook, so that DLL shouldn't be necessary. It works fine here on 64bit windows(Win7 x64 and server 2008r2). All machines are using cat 11.4 34) (Message 47321) Posted 9 Apr 2011 by Crunch3r Post: The ATI app is using nearly a full cpu core per wu....ouch! Not here, only 0.05CPU. That's what it says, but it's busy waiting for the GPU to finish things which I thought I fixed months ago. And it's ~25% slower compared to the old app... that sucks :(What GPU? I was running it for a few days in the last week and it was pretty much the same for my 5870 on a 4850 @ 760MHz it takes 249 sec compared to 188 sec using the old app (same WUs of course). On a 5870 @ 960MHz it takes 105 sec now, while the old app only took like 80 sec or so.There are different sized tasks out there, so you need to be sure to compare ones from the same set of workunits. When I was using it there were some that were around ~80s and some ~110s for both. I know, that's why i posted "same WUs" (10_3s_fix20) those that grant 367 credits. It is definetely quite a bit slower. EDIT: Are you using the math functions included with cal++ (include\cal\il\math) or something else ? 35) (Message 47316) Posted 9 Apr 2011 by Crunch3r Post: The ATI app is using nearly a full cpu core per wu....ouch! Not here, only 0.05CPU. That's what it says, but it's busy waiting for the GPU to finish things which I thought I fixed months ago. And it's ~25% slower compared to the old app... that sucks :(What GPU? I was running it for a few days in the last week and it was pretty much the same for my 5870 on a 4850 @ 760MHz it takes 249 sec compared to 188 sec using the old app (same WUs of course). On a 5870 @ 960MHz it takes 105 sec now, while the old app only took like 80 sec or so. 36) (Message 47311) Posted 9 Apr 2011 by Crunch3r Post: The ATI app is using nearly a full cpu core per wu....ouch! Not here, only 0.05CPU. That's what it says, but it's busy waiting for the GPU to finish things which I thought I fixed months ago. And it's ~25% slower compared to the old app... that sucks :( 37) (Message 47283) Posted 9 Apr 2011 by Crunch3r Post: I am getting download errors because the application to run these work units, "milkyway_0.57_windows_intelx86__ati14.exe", is not found on the server. It looks like when the server got brought back up this morning the download directory was restored to an earlier version which didn't have Matt's application added to it yet. So as soon as he wakes back up and updates the applications I'll start sending out work again (which should fix the problem). you should take a look there -> http://milkyway.cs.rpi.edu/milkyway/download/tests/ 38) (Message 47196) Posted 9 Apr 2011 by Crunch3r Post: No. I don't really see the point of checkpointing when the entire thing takes less time than the default checkpointing interval (although I'm not really sure how long it takes on the slower GPUs. I'm guessing it might take 30minutes or so on some of the lower end and older stuff?) It wouldn't be hard to add if people want it. It would be really nice if it had checkpoints implemented. (Besides, the 'low end stuff ('38xx'), which should be in the ballpark of 30min or more) there are other ATI enabled projects as well and boinc tends to switch between them in a complete irational/retareded way, sometimes suspending a WU that's ~98% done... so there you go, we need checkpoints. :) The 2.4 SDK release notes say CAL will be deprecated in the next release, so I'll make another attempt at getting the OpenCL one as fast. There was some kind of mention on the AMD forum that they were looking into the possibility of adding inline IL in CL kernels, which would be fine. The only thing then we would lose is support for 38xx GPUs. I just installed the 2.4 SDK today and havent had a closer look at it at all... anyway, just did some tests using the SKA and there where only a 2% speedup for the kernels i've tested. I'm not convinced that openCL is ready to replace CAL performance wise in the near future. Just look at some sameples from the cal++ amd compare them to OCL. Some of those are 20x faster using cal++ (Pyrit comes to mind too...) 39) (Message 47194) Posted 9 Apr 2011 by Crunch3r Post: Matt A is updating the ATI application so they'll be available and automatically downloaded when new work is generated, but for people with customized app info and who are using old versions of the ATI application this is fair warning. We should probably be generating new work sometime this weekend, and the workunits being sent out will break with the older versions of the ATI applications. The questions is how does the 'new' app perform copmpared to the old one(Gipsel) ? Does it come even close performance wise or is it nowhere near that ? Some realistic numbers would be nice to see if it'S worth running at all. Thanks.The main integral piece which matters seems to be about the same. The end likelihood I don't have on the GPU so that adds about 2-3 seconds at the end. OK that sounds good to me (especially since it's NOT openCL)! Does it checkpoint ? 40) (Message 47192) Posted 9 Apr 2011 by Crunch3r Post: Matt A is updating the ATI application so they'll be available and automatically downloaded when new work is generated, but for people with customized app info and who are using old versions of the ATI application this is fair warning. We should probably be generating new work sometime this weekend, and the workunits being sent out will break with the older versions of the ATI applications. The questions is how does the 'new' app perform copmpared to the old one(Gipsel) ? Does it come even close performance wise or is it nowhere near that ? Some realistic numbers would be nice to see if it'S worth running at all. Thanks.

Previous 20 · Next 20